Anthropic has published a new piece by Claude Code’s head of product on how the company is evolving is product development processes.

AI models are improving so fast that the traditional PM playbook (gather requirements, lock a roadmap, execute over months) is broken.

Here are the 4 new operating principles adopted by Anthropic’s product teams in this new era:

1. Plan in short sprints - ditch the long-term roadmap in favour of "side quests" - short, self-directed experiments anyone on the team can run. An afternoon prototyping something you assumed was out of reach. Some of Anthropic's most popular features - Claude Code on Desktop, the AskUserQuestion tool, and todo lists - emerged this way.

2. Demos and evals over docs - Stop writing specs first, start building first. Because you can prototype in an afternoon, wrong bets are cheap. A rough prototype anchors the conversation better than a PRD. Evals serve the same grounding function - they make abstract agent behaviour measurable and improvable.

3. Revisit features with new models - Every model release is a prompt to re-examine what you've already shipped. The best way to catch these moments is to be a daily active user and deliberately ask it to do things you think might be too hard.

4. Do the simple thing If your product works around a model limitation, that workaround becomes dead weight when the next model drops. The simpler your implementation, the easier it is to swap in new capabilities when they arrive.

Read the full post here: claude.com/blog/product…